feat: 定义不进行失败计数的错误类型 (#159)

* solve Fix: resource has been exhausted counted as failure count

* solve Fix: resource has been exhausted counted as failure count

* Update provider.go

* Update provider.go

* Update provider.go

* fix

* fix2

* fix

---------

Co-authored-by: kikii16 <[email protected]>

Author: vickyyd

internal/keypool/provider.go

@@ -73,7 +73,7 @@ func (p *KeyProvider) SelectKey(groupID uint) (*models.APIKey, error) {
 }
 
 // UpdateStatus 异步地提交一个 Key 状态更新任务。
-func (p *KeyProvider) UpdateStatus(apiKey *models.APIKey, group *models.Group, isSuccess bool) {
+func (p *KeyProvider) UpdateStatus(apiKey *models.APIKey, group *models.Group, isSuccess bool, errorMessage string) {
 	go func() {
 		keyHashKey := fmt.Sprintf("key:%d", apiKey.ID)
 		activeKeysListKey := fmt.Sprintf("group:%d:active_keys", group.ID)
@@ -83,7 +83,7 @@ func (p *KeyProvider) UpdateStatus(apiKey *models.APIKey, group *models.Group, i
 				logrus.WithFields(logrus.Fields{"keyID": apiKey.ID, "error": err}).Error("Failed to handle key success")
 			}
 		} else {
-			if err := p.handleFailure(apiKey, group, keyHashKey, activeKeysListKey); err != nil {
+			if err := p.handleFailure(apiKey, group, keyHashKey, activeKeysListKey, errorMessage); err != nil {
 				logrus.WithFields(logrus.Fields{"keyID": apiKey.ID, "error": err}).Error("Failed to handle key failure")
 			}
 		}
@@ -117,6 +117,29 @@ func (p *KeyProvider) executeTransactionWithRetry(operation func(tx *gorm.DB) er
 	return err
 }
 
+// shouldCountFailure 判断错误是否应该计入失败次数
+func (p *KeyProvider) shouldCountFailure(errorMsg string) bool {
+	if errorMsg == "" {
+		return true // 没有错误信息时默认计数
+	}
+
+	// 转换为小写进行匹配
+	errorLower := strings.ToLower(errorMsg)
+
+	// 不计入失败次数的错误模式
+	excludePatterns := []string{
+		"resource has been exhausted (e.g. check quota).", // Resource has been exhausted (e.g. check quota).
+	}
+
+	for _, pattern := range excludePatterns {
+		if strings.Contains(errorLower, pattern) {
+			return false
+		}
+	}
+
+	return true // 其他错误计入失败次数
+}
+
 func (p *KeyProvider) handleSuccess(keyID uint, keyHashKey, activeKeysListKey string) error {
 	keyDetails, err := p.store.HGetAll(keyHashKey)
 	if err != nil {
@@ -163,7 +186,7 @@ func (p *KeyProvider) handleSuccess(keyID uint, keyHashKey, activeKeysListKey st
 	})
 }
 
-func (p *KeyProvider) handleFailure(apiKey *models.APIKey, group *models.Group, keyHashKey, activeKeysListKey string) error {
+func (p *KeyProvider) handleFailure(apiKey *models.APIKey, group *models.Group, keyHashKey, activeKeysListKey string, errorMessage string) error {
 	keyDetails, err := p.store.HGetAll(keyHashKey)
 	if err != nil {
 		return fmt.Errorf("failed to get key details from store: %w", err)
@@ -175,6 +198,16 @@ func (p *KeyProvider) handleFailure(apiKey *models.APIKey, group *models.Group,
 
 	failureCount, _ := strconv.ParseInt(keyDetails["failure_count"], 10, 64)
 
+	// 判断是否应该计入失败次数
+	shouldCount := p.shouldCountFailure(errorMessage)
+	if !shouldCount {
+		logrus.WithFields(logrus.Fields{
+			"keyID":       apiKey.ID,
+			"errorMessage": errorMessage,
+		}).Debug("Error not counted towards failure threshold")
+		return nil // 不计入失败次数，直接返回
+	}
+
 	// 获取该分组的有效配置
 	blacklistThreshold := group.EffectiveConfig.BlacklistThreshold
 

internal/keypool/validator.go

@@ -61,7 +61,11 @@ func (s *KeyValidator) ValidateSingleKey(key *models.APIKey, group *models.Group
 
 	isValid, validationErr := ch.ValidateKey(ctx, key, group)
 
-	s.keypoolProvider.UpdateStatus(key, group, isValid)
+	var errorMsg string
+	if !isValid && validationErr != nil {
+		errorMsg = validationErr.Error()
+	}
+	s.keypoolProvider.UpdateStatus(key, group, isValid, errorMsg)
 
 	if !isValid {
 		logrus.WithFields(logrus.Fields{

internal/proxy/server.go

@@ -195,15 +195,14 @@ func (ps *ProxyServer) executeRequestWithRetry(
 			return
 		}
 
-		ps.keyProvider.UpdateStatus(apiKey, group, false)
-
 		var statusCode int
 		var errorMessage string
 		var parsedError string
 
 		if err != nil {
 			statusCode = 500
 			errorMessage = err.Error()
+			parsedError = errorMessage // 网络错误直接使用原始错误信息
 			logrus.Debugf("Request failed (attempt %d/%d) for key %s: %v", retryCount+1, cfg.MaxRetries, utils.MaskAPIKey(apiKey.KeyValue), err)
 		} else {
 			// HTTP-level error (status >= 400)
@@ -220,6 +219,9 @@ func (ps *ProxyServer) executeRequestWithRetry(
 			logrus.Debugf("Request failed with status %d (attempt %d/%d) for key %s. Parsed Error: %s", statusCode, retryCount+1, cfg.MaxRetries, utils.MaskAPIKey(apiKey.KeyValue), parsedError)
 		}
 
+		// 使用解析后的错误信息更新密钥状态
+		ps.keyProvider.UpdateStatus(apiKey, group, false, parsedError)
+
 		newRetryErrors := append(retryErrors, types.RetryError{
 			StatusCode:         statusCode,
 			ErrorMessage:       errorMessage,